linux-kernel - Re: [patch V4 03/15] entry: Provide generic syscall exit function

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CALCETrVwrQ6oCxwEraxLF8ia8P8HUR2czrfqYtgQEdm8DM=RLQ@mail.gmail.com>
Date:   Mon, 27 Jul 2020 15:37:01 -0700
From:   Andy Lutomirski <luto@...nel.org>
To:     Thomas Gleixner <tglx@...utronix.de>
Cc:     LKML <linux-kernel@...r.kernel.org>, X86 ML <x86@...nel.org>,
        linux-arch <linux-arch@...r.kernel.org>,
        Will Deacon <will@...nel.org>, Arnd Bergmann <arnd@...db.de>,
        Mark Rutland <mark.rutland@....com>,
        Kees Cook <keescook@...omium.org>,
        Keno Fischer <keno@...iacomputing.com>,
        Paolo Bonzini <pbonzini@...hat.com>,
        kvm list <kvm@...r.kernel.org>,
        Gabriel Krisman Bertazi <krisman@...labora.com>
Subject: Re: [patch V4 03/15] entry: Provide generic syscall exit function

On Tue, Jul 21, 2020 at 4:08 AM Thomas Gleixner <tglx@...utronix.de> wrote:
>
> Like syscall entry all architectures have similar and pointlessly different
> code to handle pending work before returning from a syscall to user space.
>
>   1) One-time syscall exit work:
>       - rseq syscall exit
>       - audit
>       - syscall tracing
>       - tracehook (single stepping)
>
>   2) Preparatory work
>       - Exit to user mode loop (common TIF handling).
>       - Architecture specific one time work arch_exit_to_user_mode_prepare()
>       - Address limit and lockdep checks
>
>   3) Final transition (lockdep, tracing, context tracking, RCU). Invokes
>      arch_exit_to_user_mode() to handle e.g. speculation mitigations
>
> Provide a generic version based on the x86 code which has all the RCU and
> instrumentation protections right.
>
> Provide a variant for interrupt return to user mode as well which shares
> the above #2 and #3 work items.

I still don't love making the syscall exit path also do the
non-syscall stuff.  Do you like my suggestion of instead having a
generic function to do the syscall complete with all the entry and
exit stuff?

The singlestep handling is a mess.  I'm not convinced that x86 does
this sensibly.  Right now, I *think* we are quite likely to not send
SIGTRAP on the way out of syscalls if TF is set, and we'll actually
execute one more user instruction before sending the signal.  One
might reasonably debate whether this is a bug, but we should probably
figure it out at some point.

That latter bit is relevant to your patch because the fix might end up
being something like this:

void do_syscall_64(...)
{
  unsigned long orig_flags;
  idtentry_enter();
  instrumentation_begin();
  generic_do_syscall(regs, regs->orig_ax, AUDIT_ARCH_X86_64);
  if (unlikely(orig_flags & regs->flags & X86_EFLAGS_TF))
    raise SIGTRAP -- pretend we got #DB.
  instrumentation_end();
  idtentry_exit(); <-- signal is delivered here
}

That logic is probably all kinds of buggy, but the point is that the
special handling probably wants to be done between the generic syscall
code and the exit code.